KAT-Coder V2.5-Dev
Lightweight agentic coding.轻量级智能体编程。
- 35B total · 3B active parameters35B 总参数 · 3B 激活参数
- Open model weights开放模型权重
- 3M+ model & community downloads原模型与社区量化版累计下载超 300 万
Learning to reason.
Learning to create.
Understanding the world.
学习推理，学习创造，理解世界。
Exploring intelligent systems through language, generative modeling, and the physics of a changing world.从语言、生成式建模与物理规律出发，探索智能系统。
Latest LLM Releases大语言模型最新发布
KAT-Coder V2.5
Coding at trillion-parameter scale.万亿参数规模的编程模型。
- Reproducible coding environments可复现的编程环境
- Specialist distillation for long tasks专家蒸馏，面向长程任务
- Competitive reported coding results论文报告编程表现接近 GLM-5.2 与 Claude Opus 4.8
KAT-Coder V2
Specialize first. Then unify.先专精，再统一。
- Five domain specialists五类领域专家
- Coding and tool-use capabilities编程与工具使用能力
- Unified across agent frameworks统一跨智能体框架能力
Large Language Models大语言模型
Generative AI生成式人工智能
Restoration & Super-Resolution复原与超分辨率
ShiftLUT: Spatial Shift Enhanced Look-Up Tables for Efficient Image Restoration让高效图像恢复看得更远
Plug-and-Play Tri-Branch Invertible Block for Image Rescaling重新思考小图像应当保留的信息
OAPT: Offset-Aware Partition Transformer for Double JPEG Artifacts Removal恢复经过二次 JPEG 压缩的图像
CPGA: Coding Priors-Guided Aggregation Network for Compressed Video Quality Enhancement利用编解码器自身的线索恢复视频
XPSR: Cross-modal Priors for Diffusion-based Image Super-Resolution利用语言引导忠实的超分辨率重建
PTM-VQA: Efficient Video Quality Assessment Leveraging Diverse PreTrained Models from the Wild从多种预训练模型中学习视频质量
CasSR: Activating Image Power for Real-World Image Super-Resolution为生成式超分辨率提供更清晰的参考
Physics & Fluid Dynamics物理与流体动力学
